Update README.md
Browse files
README.md
CHANGED
@@ -4,10 +4,52 @@ language:
|
|
4 |
- en
|
5 |
pipeline_tag: text-generation
|
6 |
base_model: EleutherAI/pythia-31m
|
|
|
|
|
|
|
|
|
7 |
---
|
8 |
|
9 |
# pythia-31m (fp32)
|
10 |
|
11 |
|
12 |
|
13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
- en
|
5 |
pipeline_tag: text-generation
|
6 |
base_model: EleutherAI/pythia-31m
|
7 |
+
datasets:
|
8 |
+
- EleutherAI/pile
|
9 |
+
tags:
|
10 |
+
- smol
|
11 |
---
|
12 |
|
13 |
# pythia-31m (fp32)
|
14 |
|
15 |
|
16 |
|
17 |
+
This is [EleutherAI/pythia-31m](https://huggingface.co/EleutherAI/pythia-31m) but saved explicitly in fp32 - see safetensors params. It is smaller than the other 'official' checkpoints included in the Pythia study.
|
18 |
+
|
19 |
+
|
20 |
+
|
21 |
+
|
22 |
+
## config/info
|
23 |
+
|
24 |
+
|
25 |
+
```json
|
26 |
+
{
|
27 |
+
"_name_or_path": "EleutherAI/pythia-31m",
|
28 |
+
"architectures": [
|
29 |
+
"GPTNeoXForCausalLM"
|
30 |
+
],
|
31 |
+
"attention_dropout": 0.0,
|
32 |
+
"bos_token_id": 0,
|
33 |
+
"classifier_dropout": 0.1,
|
34 |
+
"eos_token_id": 0,
|
35 |
+
"hidden_act": "gelu",
|
36 |
+
"hidden_dropout": 0.0,
|
37 |
+
"hidden_size": 256,
|
38 |
+
"initializer_range": 0.02,
|
39 |
+
"intermediate_size": 1024,
|
40 |
+
"layer_norm_eps": 1e-05,
|
41 |
+
"max_position_embeddings": 2048,
|
42 |
+
"model_type": "gpt_neox",
|
43 |
+
"num_attention_heads": 8,
|
44 |
+
"num_hidden_layers": 6,
|
45 |
+
"rope_scaling": null,
|
46 |
+
"rotary_emb_base": 10000,
|
47 |
+
"rotary_pct": 0.25,
|
48 |
+
"tie_word_embeddings": false,
|
49 |
+
"torch_dtype": "float32",
|
50 |
+
"transformers_version": "4.33.1",
|
51 |
+
"use_cache": true,
|
52 |
+
"use_parallel_residual": true,
|
53 |
+
"vocab_size": 50304
|
54 |
+
}
|
55 |
+
```
|